Methods and apparatus compress data, comprising an In-phase (I) component and a Quadrature (Q) component. statistical characteristics of the data are utilized to convert the data into a form that requires fewer bits in accordance with the statistical characteristics. The data may be further compressed by transforming the data and by modifying the transformed data in accordance with a quantization conversion table that is associated with the processed data. Additionally, redundancy may be removed from the processed data with an encoder. Subsequent processing of the compressed data may decompress the compressed data in order to approximate the original data by reversing the process for compressing the data with corresponding inverse operations. interleaved I and Q components can be processed rather than separating the components before processing the data. The processed data type may be determined by providing metadata to retrieve the appropriate quantization table from a knowledge database.
|
15. An apparatus for decompressing data, the apparatus comprising:
a decompression module that obtains a first compressed component and a second compressed component, that converts the first compressed component to a first decompressed component in accordance with a statistical characteristic of the data, and that interleaves a first plurality of first compressed component samples with a second plurality of second compressed component samples.
1. A method for compressing received data comprising a first received component and a second received component, the method comprising:
(a) receiving an interleaved stream of a first plurality of first received component samples with a second plurality of second received component samples;
(b) converting the first received component to a first converted component in accordance with a statistical characteristic; and
(c) converting the second received component to a second converted component in accordance with the statistical characteristic.
8. A method for decompressing data in order to approximate original data, the original data comprising a first original component and second original component, the method comprising:
(a) receiving an interleaved stream of a first converted plurality of first converted component samples and a second converted plurality of second converted component samples;
(b) obtaining a first converted component and a second converted component; and
(c) converting the first converted component to a first decompressed component in accordance with a statistical characteristic.
2. The method of
(d) determining the statistical characteristic that is associated with the first received component and the second received component.
3. The method of
4. The method of
(d) transforming the first converted component into a first transformed component and the second converted component into a second transformed component, wherein a first transformed plurality of first transformed component samples is interleaved with a second transformed plurality of second transformed component samples;
(e) interleaving a first transformed plurality of first transformed component samples with a second transformed plurality of second transformed component samples;
(f) quantizing the first transformed component into a first quantized transform and the second transformed component into a second quantized transform; and
(g) interleaving a first quantized plurality of first quantized transform samples with a second quantized plurality of second quantized transform samples.
5. The method of
(h) encoding the first quantized transform into a first compressed component and the second quantized transform into a second compressed component; and
(i) interleaving a first compressed plurality of first compressed component samples with a second compressed plurality of second compressed component samples.
6. The method of
In-phase (I) component and the second received component corresponds to a received Quadrature (Q) component.
7. A computer-readable medium having computer-executable instructions that, when executed by a computer, cause the computer to perform steps recited in
9. The method of
(d) determining the statistical characteristic that is associated with the first original component.
10. The method of
11. The method of
12. The method of
(i) obtaining a first quantized transform and a second quantized transform;
(ii) inverse quantizing the first quantized transform into a first transformed component and the second quantized transform into a second transformed component; and
(iii) inverse transforming the first transformed component into the first converted component and the second transformed component into the second converted component.
13. The method of
(1) obtaining a first compressed component and a second compressed component; and
(2) decoding the first compressed component into the first quantized transform and the second compressed component into the second quantized transform.
14. A computer-readable medium having computer-executable instructions that, when executed by a computer, cause the computer to perform steps recited in
|
This application is a continuation application of U.S. application Ser. No. 10/776,310 of commonly-owned U.S. Pat. No. 7,136,010 filed on Feb. 11, 2004, and issued on Nov. 14, 2006, which is a continuation-in-part of Ser. No. 10/269,818 common-owned, U.S. Pat. No. 6,714,154 filed on Oct. 11, 2002, and issued on Mar. 30, 2004, and claiming priority to provisional U.S. Application No. 60/392,316, filed Jun. 28, 2002, the contents of each being bodily incorporated by references in its entirety.
The present invention relates to compressing and decompressing data such as synthetic aperture radar data.
Compression of Synthetic Aperture Radar (SAR) data may require that both magnitude and phase information be preserved.
Video phase history data 104 are then passed through a Phase History Processor (PHP) 105 where data 104 are focused in both range (corresponding to a range focusing apparatus 107) and azimuth (corresponding to an azimuth focusing apparatus 109). The output of phase history processor 105 is referred to as Single Look Complex (SLC) data 110. A detection function 111 processes SLC data 110 to form a detected image 112.
Existing complex SAR sensors collect increasingly large amounts of data. Processing the complex data information and generating resultant imagery products may utilize four to eight times the memory storage and bandwidth that is required for the detected data (I&Q). In fact, some studies suggest exponential growth in associated data throughput over the next decade. However, sensors are typically associated with on-board processors that have limited processing and storage capabilities. Moreover, collected data are often transmitted to ground stations over a radio channel having a limited frequency bandwidth. Consequently, collected data may require compression in order to store or transmit collected data within resource capabilities of data collecting apparatus. Also, a SAR compression algorithm should be robust enough to compress both VPH data 104 and SLC SAR data 110, should produce visually near-lossless magnitude image, and should cause minimal degradation in resultant products 112.
Several compression algorithms have been proposed to compress SAR data. However, while such compression algorithms generally work quite well for magnitude imagery, the compression algorithms may not efficiently compress phase information. Moreover, the phase component may be more important in carrying information about a SAR signal than the magnitude component. With SAR data 102, compression algorithms typically do not achieve compression ratios of more than ten to one without significant degradation of the phase information. Because many of the compression algorithms are typically designed for Electro/Optical (EO) imagery, the compression algorithms rely on high local data correlation to achieve good compression results and typically discard phase data prior to compression. Table 1 lists several compression algorithms discussed in the literature and provide a brief description of each.
TABLE 1
Popular Alternative SAR Data Compression Algorithms
Compression Algorithm
Description
Block Adaptive Quantization
Choice of onboard data compression
(BAQ)
methods due to simplicity in coding and
decoding hardware. Low compression
ratios achieved (<4:1).
Vector Quantization (VQ)
Codebook created assigning a number
for a sequence of pixels. Awkward
implementation since considerable
complexity required in codebook
formulation.
Block Adaptive Vector
Consists of first compressing data with
Quantization (BAVQ)
BAQ and then following up with VQ.
Similar to BAQ.
Karhunen-Loeve Transform
Statistically optimal transform for
(KLT)
providing uncorrelated coefficients;
however, computational cost is large.
Fast Fourier Transform
2-D Fast Fourier Transform (FFT)
BAQ (FFT-BAQ)
performed on raw SAR data. Before
raw data is transformed, dynamic
range for each block is decreased
using a BAQ.
Uniform Sampled Quantization
Emphasizes phase accuracy of selected
(USQ)
points.
Flexible BAQ (FBAQ)
Based on minimizing mean square error
between original and reconstructed data.
Trellis-Coded Quantization
Unique quantizer optimization design.
(TCQ)
Techniques provide superior signal
to noise ratio (SNR) performance
to BAQ and VQ for SAR.
Block Adaptive Scalar
BSAQ's adaptive technique provides
Quantization (BSAQ)
some performance improvement.
Existing optical algorithms are inadequate for compressing complex multi-dimensional data, such as SAR data compression. For example with optical imagery, because of a human eyesight's natural high frequency roll-off, the high frequencies play a less important role than low frequencies. Also, optical imagery has high local correlation and the magnitude component is typically more important than the phase component. However, such characteristics may not be applicable to complex multi-dimensional data. Consequently, a method and apparatus that provides a large degree of compression without a significant degradation of the processed signal are beneficial in advancing the art in storing and transmitting complex multi-dimensional data. Furthermore, the quality of the processed complex multi-dimensional data is not typically visually assessable. Thus, a means for evaluating the effects of compression on the resulting processed signal is beneficial to adjusting and to evaluating the compression process.
The present invention provides methods and apparatus for compressing data comprising an In-phase (I) component and a Quadrature (Q) component. The compressed data may be saved into a memory or may be transmitted to a remote location for subsequent processing or storage. Statistical characteristics of the data are utilized to convert the data into a form that requires a reduced number of bits in accordance with its statistical characteristics. The data may be further compressed by transforming the data, as with a discrete cosine transform, and by modifying the transformed data in accordance with a quantization conversion table that is selected using a data type associated with the data. Additionally, a degree of redundancy may be removed from the processed data with an encoder. Subsequent processing of the compressed data may decompress the compressed data in order to approximate the original data by reversing the process for compressing the data with corresponding inverse operations.
In a first embodiment of the invention, data are compressed with an apparatus comprising a preprocessor, a transform module, a quantizer, an encoder, and a post-processor. The preprocessor separates the data into an I component and a Q component and bins each component according to statistical characteristics of the data. The transform module transforms the processed data into a discrete cosine transform that is quantized by the quantizer using a selected quantization conversion table. The encoder partially removes redundancy from the output of the quantizer using Huffman coding. The resulting data can be formatted by a post-processor for storage or transmittal. With a second embodiment, the preprocessor converts the I and Q components into amplitude and phase components and forms converted I and Q components.
Variations of the embodiment may use a subset of the apparatus modules of the first or the second embodiment. In a variation of the embodiment, the apparatus comprises a preprocessor, a transform module, and a quantizer.
With another embodiment of the invention, interleaved I and Q components of SAR data can be processed rather than separating the components before processing the data. Thus, one is not constrained to separately compress and decompress the I and Q data. With the processing of interleaved I and Q data, one may assume that the statistical characteristics are the same, in which a deviation from this assumption may result in a reduced compression ratio. Moreover, the processing of interleaved data is applicable to data that are characterized by one or more dimensions.
With another embodiment of the invention, the data type being processed for compression and decompression is determined. With the compression process, the preprocessor interacts with the quantizer. The preprocessor provides metadata to the quantizer so that the quantizer can retrieve the appropriate quantization table from a knowledge database. The metadata may include statistical characteristics of the data being processed such as the mean, the standard deviation, and the maximum/minimum values. This aspect of the invention also supports a training mode, in which the quantizer informs the preprocessor so that a new data type can be specified.
A more complete understanding of the present invention and the advantages thereof may be acquired by referring to the following description in consideration of the accompanying drawings, in which like reference numbers indicate like features and wherein:
In the following description of the various embodiments, reference is made to the accompanying drawings which form a part hereof, and in which is shown by way of illustration various embodiments in which the invention may be practiced. It is to be understood that other embodiments may be utilized and structural and functional modifications may be made without departing from the scope of the present invention.
Other embodiments of the invention may support other applications of complex multidimensional data, including weather data, oil and gas exploration data, encrypted/decrypted data, medical archival of MRI/CTI and three dimensional sonograms, digital video signals, and modem applications.
Referring again to
M=(I2+Q2)1/2 (EQ. 1)
φ=tan−1(Q/I) (EQ. 2)
Moreover, I and Q components may be obtained from the magnitude and the phase components by using Equations 3 and 4.
I=M cos φ (EQ. 3)
Q=M sin φ (EQ. 4)
Additionally, the power of a SAR signal may provide good visual results when printing intensity (magnitude-only) imagery. The power of a SAR signal may be obtained from Equation 5.
P=20 log10M2 (EQ. 5)
The conversion between (I, Q) and (M, φ) as expressed in EQs. 1-4 allows SAR data 202 to be studied in both data formats before and after compression. When SAR data 202 are represented as magnitude and phase components, additional bits may be allocated to the phase component versus the magnitude component to achieve the least degradation of the phase product, depending on characteristics of SAR data 202. In an embodiment, more bits (e.g. six bits) of the phase component and fewer bits (e.g. two bits) of the magnitude component are used to generate compressed I and Q components. Conversely, when SAR data 202 are represented by in-phase and quadrature components, apparatus 200 can process the in-phase component separately from the quadrature component for a single complex image.
Preprocessor 201 also determines a data type (as discussed in the context of
An adaptive source data calculations module 301 separately processes the I and Q components of (I,Q) data pairs 202 in order to determine corresponding statistical characteristics. (An example of statistical characteristics is shown in
In a variation of the embodiment, Single Look Complex (SLC) data 110 are transformed using a Fast Fourier Transform (FFT) prior to binning data 202 by modules 303 and 305, wherein a transformation of SLC data 110 has statistical characteristics that are similar to VPH data 104. (In the embodiment, modules 303 and 305 bin data 202 by first processing the I component and subsequently processing the Q component.) However, other embodiments of the invention may utilize other transform types in order to modify statistical characteristics of the data. After quantization by modules 303 and 305, the transformed SLC data are inversely transformed using an Inverse Fast Fourier Transform (IFFT).
Apparatus 200 may use the same statistical modeling for the In-phase (I) and Quadrature (Q) components if both components have approximately the same statistical characteristics.
Preprocessor 201 accommodates different sensor types regarding a data format and a number of bits per pixel. (A pixel corresponds to a point in the corresponding image being scanned by a radar system.) SAR data 202 are typically 64 bits (with 32 bits for the I component and 32 bits for the Q component for each pixel) or 32 bits (with 16 bits for the I component and 16 bits for the Q component for each pixel). Preprocessor 201 determines the range of pixel values and the best bin assignment. Values of the I and Q components are converted to 8-bit formats with more bits being allocated from the associated phase component than the magnitude component before reducing the I and Q components to 8-bit formats. (As previously discussed, two bits from the magnitude component and six bits from the phase component, as determined from Equations 1 and 2 by converting I and Q into M and φ, are used to generate compressed components I′ and Q′, as determined from Equations 3 and 4 by converting the truncations of M and φ into I′ and Q′.)
Processed SAR data 204 (comprising a converted I component and a converted Q component) are further processed through transform module 203 using a Discrete Cosine Transform (DCT) in order to obtain the frequency representation of the in-phase and the quadrature data as transformed data 206 (comprising a transformed I component and a transformed Q component). As will be discussed in the context of
In Equation 6, pair (i,j) represents a pixel of processed SAR data 204 within a block (which is a portion, A(i,j) represents a corresponding in-phase or quadrature value of the pixel, and B(k1,k2) represents a corresponding DCT coefficient, where pair (k1,k2) identifies the DCT coefficient in the DCT matrix. In the embodiment, a DCT coefficient is calculated over an eight by eight pixel block, i.e. N1 and N2 equal 8, although other embodiments of the invention may use a different value for N. (The collection of DCT coefficients may be represented by an 8 by 8 matrix.)
Each element of a DCT matrix (e.g. matrix 703) is arithmetically divided by a corresponding element of the quantization matrix and rounded to an integer, thus providing quantized DCT data 208 (comprising a quantized I transform or a quantized Q transform). Each element of the quantization matrix is determined by statistics for the corresponding DCT coefficient in accordance with a specified maximum error (e.g. a root mean square error, a peak signal to noise ratio, and a byte by byte file comparison). (
In a variation of the embodiment, the quantization matrix may be determined by reducing a Measurement and Signature Intelligence (MASINT) product distortion. (In some cases, the reduction may correspond to a minimization of the distortion.) The distortion may be determined from interferometric SAR, coherent change detection (CCD), and polarimetry products. Interferometric SAR (IFSAR) is a comparison of two or more coherent SAR images collected at slightly different geometries. The process extracts phase differences caused by changes in elevation within the scene. IFSAR produces digital terrain elevation data suitable for use in providing terrain visualization products. (Products are generally referred as Digital Elevation Models (DEM).) These products are used in mapping and terrain visualization products. The advantage of IFSAR height determination is that is much more accurate than other methods, such as photo/radargrammetry methods that use only the intensity (magnitude) data, because phase is used and height determination is done with wavelength measurements which are very accurate (i.e. for commercial systems at C Band (5 GHz) approximately 5.3 cm)).
Coherent Change Detection (CCD) is a technique involving the collection and comparison of a registered pair of coherent SAR images from approximately the same geometry collected at two different times (before and after an event). The phase information, not the magnitude, is used to determine what has changed between the first and second collection. This can determine scene changes to the order of a wavelength (5.3 cm) and may denote ground changes/activity occurring between collections.
Polarimetry products are generally collected using systems that can independently radiate and collect vertical and horizontal complex SAR data. This technique is accomplished by alternately radiating vertical and horizontally polarized SAR pulses, receiving on both horizontal and vertical antennas, and saving the complex data from each. The product formed is a unique target signature for objects with an associated complex polarized radar reflectance. This technique is used in many automatic target recognition systems (ATR).
In a variation of the embodiment, each member of the quantization matrix (associated with a quantization conversion table) is determined by a heuristic process 1100 as shown in
In another variation of the embodiment, the quantization matrix is determined by the statistical characteristics of the DCT matrix, as was previously discussed. The quantization matrix is subsequently modified according to heuristic process 1100.
Transformed data 206 are quantized according to corresponding transform statistics that are associated with the DCT coefficients. DCT coefficients can be represented as departures from a standard statistical distribution function (e.g., Laplacian, Gaussian, or Rayleigh). (A Laplacian function has a form of e−|x|, while a Gaussian function has a form of e−x
Other embodiments of the invention may utilize other transform types such as a Discrete Fourier Transform (DFT) or a discrete z-transform, both transforms being well known in the art. However, with a selection of a different transform, the transform statistics may be different as reflected by the design of quantizer 205.
Quantized SAR data 208 are consequently processed by encoder 207 (e.g. a Huffman encoder). Each output 210 (comprising a compressed I component and a compressed Q component) of encoder 207 comprises an encoder pair (comprising a number of zeros that precede output 210 and a number of bits that represent a value of the corresponding DCT coefficient) and the value of the corresponding DCT coefficient (SAR data 208). Encoder 207 may provide additional compression by removing a degree of redundancy that is associated with the encoder pair and SAR data 208 (in which frequently occurring data strings that are associated with the quantized DCT coefficients are replaced with shorter codes). Other embodiments of the invention may utilize other types of encoders such as Shannon Fano coding and Arithmetic coding. Encoder 207 provides encoded data 210 to post-processor 209.
Post-processor 209 may further process encoded data 210 in order to format data 210 into a format that is required for storing (that may be associated with archiving compressed data) or for transmitting compressed data 212 through a communications device. In the embodiment, the communications device may be a radio frequency transmitter that transmits from a plane to a monitoring station, utilizing a radio data protocol as is known in the art. In the embodiment, for example, post-processor 209 may format a data file (corresponding to a SAR image) into records that can be accommodated by a storage device. Also, post-processor may include statistical information and the data type regarding SAR data 202. The statistical information and the data type may be used for decompressing compressed SAR data 212 at a subsequent time.
Compressed data 212 may be subsequently decompressed by using apparatus that utilizes inverse operations corresponding to the operations that are provided by apparatus 200 in a reverse order.
In the embodiment, compressed data 212 may be compliant with National Imagery Transmission Format (NITF) standards, in which header information about user-defined data (e.g. a quantization matrix) may be included. Thus, compressed data 212 may be compatible with processing software in accordance with Joint Photographic Experts Group (JPEG) compression standards.
Other embodiments of the invention may compress and decompress data that are characterized by a different number of components (often referred as dimensions). Data that is characterized by more than one component (e.g. 2, 3, or more components) are often referred as multidimensional data. In such cases, preprocessor 201 may determine statistical characteristics associated with each of the components and map each of the components to bins in accordance with the statistical characteristics. Transform module 203 transforms each of the components according to a selected transform (e.g. a Fast Fourier Transform). Quantizer 205 subsequently quantizes each of the transformed components.
Classical Electro-Optical (EO) based metrics, such as root mean square error (RMSE) and Peak Signal to Noise Ratio (PSNR), are useful for evaluating the magnitude imagery, but the EO-based metrics may not provide sufficient information about the phase data or the other derived products. EO-based metrics provide a necessary but not a sufficient condition for complex data compression fidelity. Useful magnitude imagery may also be available from the compression process. The processes that generate phase data driven products such as interferometry, CCD and polarimetry may be included in the evaluations. Additional SAR data product metrics may be implemented to evaluate the phase information and any degradation of the products caused by compression.
An evaluation of compressed exemplary SAR data as processed by apparatus 200 indicates that, with SAR data 202 being compressed at ratios greater than twenty to one, apparatus 200 may achieve near-lossless results for magnitude images and minimal degradation to phase information.
Processing Interleaved Components
Referring to
Referring to
In the embodiment, modules 1601 and 1603 processes interleaved data, where the I component and the Q component have approximately the same statistical characteristics. Typically, the resulting compression ratio decreases the more that the statistical characteristics of the I component differ from those of the Q component.
Determination of Data Type
As with adaptive control loop 251 as shown in
If module 1705 cannot identify the data type from the metadata, module 1705 selects a default quantization table and uses the default quantization table from knowledge base 1711 as similarly explained in the context of
After quantizing the data, encoder module 1707 encodes the data, as previously explained with
Step 1801 performs header analysis if header information is included with SAR data 202 by reading information at the beginning of a data file that identifies the format of the data being analyzed. (The operation of step 1801 is similar to the header analysis performed by adaptive source data calculations module 301 previously explained with
If header data is not available, SAR data 202 is analyzed in order to deduce the metadata. In step 1805, the number of bits per data word is determined. If the order of bytes is reversed with respect to the assumed order that preprocessor 201 uses for processing SAR data 202, as determined by step 1807, the bytes are swapped in step 1809. (Operation of step 1807 and 1809 are similar to the operation as performed by preprocessor 201 as shown in
In step 1811, the maximum value and the minimum value of SAR data 202 are determined. Furthermore, the mean and the standard deviation of SAR data 202 are calculated in step 1813. The metadata (comprising the minimum value, the maximum data, the mean, and the standard deviation) are compared to corresponding metadata of known data types in step 1815. In step 1817, if the data type is known, the data type is identified in step 1821. Otherwise, the data type is assumed to be the default data type in step 1819.
As can be appreciated by one skilled in the art, a computer system with an associated computer-readable medium containing instructions for controlling the computer system can be utilized to implement the exemplary embodiments that are disclosed herein. The computer system may include at least one computer such as a microprocessor, digital signal processor, and associated peripheral electronic circuitry.
Poehler, Paul L., Cirillo, Francis R.
Patent | Priority | Assignee | Title |
Patent | Priority | Assignee | Title |
4801939, | Sep 04 1987 | Unisys Corporation | High-speed data compressor/decompressor for synthetic aperture radar |
5535013, | Apr 19 1991 | Matsushita Electric Industrial Co., Ltd. | Image data compression and expansion apparatus, and image area discrimination processing apparatus therefor |
5546085, | Dec 05 1994 | Lockheed Martin Corporation | Separating coherent radio-frequency interference from synthetic aperture data |
5661477, | Aug 05 1994 | Deutsche Forschungsanstalt fur Luft- und Raumfaht e.V. | Methods for compressing and decompressing raw digital SAR data and devices for executing them |
6005982, | Aug 29 1996 | Hoya Corporation | Image compression and expansion device |
7136010, | Jun 28 2002 | Leidos, Inc | Measurement and signature intelligence analysis and reduction technique |
20040021600, | |||
EP586225, | |||
GB2411087, | |||
WO2004004309, |
Executed on | Assignor | Assignee | Conveyance | Frame | Reel | Doc |
Nov 06 2006 | Science Applications International Corporation | (assignment on the face of the patent) | / | |||
Sep 27 2013 | Science Applications International Corporation | Leidos, Inc | CHANGE OF NAME SEE DOCUMENT FOR DETAILS | 032663 | /0938 | |
Aug 16 2016 | Leidos, Inc | CITIBANK, N A | SECURITY INTEREST SEE DOCUMENT FOR DETAILS | 039809 | /0801 | |
Jan 17 2020 | CITIBANK, N A , AS COLLATERAL AGENT | Leidos, Inc | RELEASE BY SECURED PARTY SEE DOCUMENT FOR DETAILS | 051632 | /0742 |
Date | Maintenance Fee Events |
Jun 13 2011 | M1551: Payment of Maintenance Fee, 4th Year, Large Entity. |
Jun 11 2015 | M1552: Payment of Maintenance Fee, 8th Year, Large Entity. |
Jul 29 2019 | REM: Maintenance Fee Reminder Mailed. |
Jan 13 2020 | EXP: Patent Expired for Failure to Pay Maintenance Fees. |
Date | Maintenance Schedule |
Dec 11 2010 | 4 years fee payment window open |
Jun 11 2011 | 6 months grace period start (w surcharge) |
Dec 11 2011 | patent expiry (for year 4) |
Dec 11 2013 | 2 years to revive unintentionally abandoned end. (for year 4) |
Dec 11 2014 | 8 years fee payment window open |
Jun 11 2015 | 6 months grace period start (w surcharge) |
Dec 11 2015 | patent expiry (for year 8) |
Dec 11 2017 | 2 years to revive unintentionally abandoned end. (for year 8) |
Dec 11 2018 | 12 years fee payment window open |
Jun 11 2019 | 6 months grace period start (w surcharge) |
Dec 11 2019 | patent expiry (for year 12) |
Dec 11 2021 | 2 years to revive unintentionally abandoned end. (for year 12) |